Huntington Bank's plan to obtain a return on gen AI

• Key Insight: Huntington Bank has tied generative AI to 10%–15% cost and revenue ROI targets.
• What's at Stake: Banks risk falling years behind competitors without measurable ROI.
• Forward Look: Agentic AI will automate multi‑person tasks; controls and baselines become essential.

Source: Bullets generated by AI with editorial review

Huntington Bank's Chief Financial Officer Zach Wasserman has made it a mission to achieve a return on the bank's generative AI investments. It's on track to achieve 10% to 15% cost reductions and 10% to 15% revenue increases through targeted uses of large language models and agentic AI, he told American Banker in an interview.

"I think leaders need to find a way to balance the urgency [of keeping up or staying ahead in the AI race] with making sure they actually have an ROI and that the technology is usable by the organization," Wasserman said.

The bank's renewed focus on return on AI comes two months after MIT released a report that found that 95% of businesses are achieving zero return on generative AI.

"That's a sobering statistic and it reflects something that we should already know: AI implementation isn't plug-and-play," Theo Lau, founder of Unconventional Ventures and author of the book "Banking on (Artificial) intelligence" wrote in a blog Wednesday. "But I'd also push back on the narrative of failure. What exactly are we measuring? And what is the time frame that we are giving ourselves to experiment and gauge the technology? Are we using the right metrics?"

Wasserman, who was put in charge of AI at Huntington Bank late last year, noted that most investments don't pay off in the early stages. "We're really trying to be focused, in a disciplined way, understanding what objectives we're trying to get out of any one of these projects, and then holding accountabilities to actually generate that return," he said.

Sumeet Chabria, CEO of consultancy ThoughtLinks, told American Banker that overall, banks are doing better than "the 95% no-ROI headline."

Many first-wave generative AI investments did miss ROI targets. "However, banks were experimenting," Chabria said. "Pilots were opportunistic, run in silos and were not integrated well into core business processes, which is a must. Banks that are careful in selection and rearchitect their workflows and processes around generative AI, and tie them to activity-level baselines to drive clear business outcomes are definitely seeing value."

Huntington has deployed generative AI for 28 use cases so far, and Wasserman aims to get to 100.

"This sounds like hyperbole, but it really is true that this will pervade almost every area," Wasserman said.

Where Huntington uses generative AI

Like many large financial institutions, Huntington has been using AI in the form of more traditional machine learning for several years, for things like pricing credit, behavioral analysis and next-best-product recommendations.

"The thing that is clearly gaining a ton of momentum right now is large language model use of generative AI," Wasserman said. "We're still very early days on that."

The gen AI use case that's farthest along at Huntington is software development. "We're seeing a very significant amount of productivity lift there," Wasserman said. "Software developers like the fact that it is helping them be more productive."

Huntington has also been using large language models and agentic AI to reengineer and streamline internal processes, for instance customer service processes that involve "lots of people and lots of handoffs, where you could simplify parts of that process with agentic tools," Wasserman said.

An example is exceptions handling, which often involve escalations to specialized teams, documentation of past interactions and research into root causes and suggested resolutions.

"Gen AI tools are perfect for this use case," Wasserman said. "They can quickly summarize the situation and past conversations in previous interactions with the customer. They can be trained to automatically trigger escalation protocols to reduce the time it takes for the next step in the process to begin. And they can assess recommended servicing solutions versus standardized policies and procedures to ensure a strong adherence to key controls."

The next biggest area of gen AI deployment is general employee productivity. In many cases, Huntington is using the embedded gen AI copilots in enterprise software.

"Salesforce is a major partner," Wasserman said. "They're doing a ton of work around AI. Microsoft is another major partner." Huntington also partners with the three major cloud providers, he said (Google Cloud, Amazon Web Services and Microsoft Azure).

About 80% of the time, the bank relies on tech partners like these, the other 20% of the time its programmers create bespoke models.

His own finance department uses gen AI for regulatory reporting and internal financial reporting, "where the teams get data from multiple sources, and they have to validate and match and make sure that they're getting good quality data and that can be made more efficient."

Personalization is another area where the bank is using gen AI. "Just in the past couple months, we've begun to do a lot of re-engineering of our customer application processes to make them more dynamic, more relevant for people," Wasserman said.

Huntington, like other sizable banks, feels the pressure to stay ahead in the AI race.

"If you think about an environment where the power of a tool is increasing exponentially, if you're not on the cutting edge of utilizing that tool in six months, you won't be six months behind, you'll be two years behind," Wasserman said. "We need to be focused very intently on understanding where this is going and plan to quickly deploy this capability."

Three sources of return on AI at Huntington

Wasserman sees three ways Huntington can generate a return on its gen AI investment: cost reductions, increased revenue and better, more interesting work for employees (which could lead to attracting and retaining talented people).

Streamlining "customer journeys," or processes affecting customers, using generative AI can bring about a 10% to 15% reduction in cost and a 10% to 15% revenue lift, he said. At the same time, the employee and customer experience will improve because of the simpler, faster processes, he said.

According to Wasserman, in the last five years, overall expenses at Huntington have grown at a 5% compound annual growth rate, or CAGR, while technology investments have grown at a 25% CAGR.

The revenue increases will come from things like faster customer acquisition, better conversion of customer acquisition funnels from the top of the funnel down through into the actual final application, and deeper engagement with customers because they'll be receiving more personalized and contextually relevant information, Wasserman said.

Agentic AI will take this all further.

Currently the bank uses agentic AI in software development, employee productivity, customer service and marketing.

"These agents will effectively be taking on jobs for the company that we used to hire people to do," Wasserman said. His team is looking at where agentic AI could be used to automate parts of jobs and parts of processes that affect multiple people.

Where some people worry about agentic AI compounding generative AI's mistakes, the way Wasserman sees it, "the output of these AI-assisted tools is fundamentally no different than the output of a human-assisted process. It's just an automated version of it. And so the control mechanisms that we need to put in place for human driven processes are similar to automated processes." The bank tests models' output to make sure it stays within the parameters its set.

The key to obtaining a return on generative AI is to directly tie it to relevant metrics, Chabria said.

"That means baselining at the activity level and tracking before and after, not just engagement, but whether that engagement drives measurable value like quicker onboarding or an acquisition," Chabria said. "For example, a generative AI feature in a mobile app that intercepts a call that would normally go to the contact center may save about $5 per offshore call or $10 per onshore call. But you need to be tracking cost per call in advance of the change."